Code reviews in open source projects : how do gender biases affect participation and outcomes?
Code reviews in open source projects : how do gender biases affect participation and outcomes?
复制标题
DOI：
10.1007/s10664-023-10324-9
复制
发表时间：
2022-09
影响因子：
4.1
通讯作者：
Sayma Sultana;Asif Kamal Turzo;Amiangshu Bosu
中科院分区：
文献类型：
--
作者：
Sayma Sultana;Asif Kamal Turzo;Amiangshu Bosu
当代软件开发组织缺乏多样性，自由和开源软件（FOSS）社区中的女性比例甚至低于行业平均水平。虽然最近的研究结果暗示存在对妇女的偏见，目前还不清楚在何种程度上这种偏见影响的结果，各种软件开发tasks. ESTA研究概念上复制最近的两项研究由特雷尔等人。和Bosu和Sultana调查性别偏见的自由/开源软件社区。我们的目标是确定代码审查（或拉取请求）的结果或参与是否受到开发人员性别的影响。特别是，我们关注两个结果方面（即，代码接受和评审间隔）和一个参与方面（即，方法以1010个自由/开源软件项目为研究对象，对1010个自由/开源软件项目的代码评审过程进行了分析。这些项目中有10个使用基于Gerrit的代码审查。剩下的1000个是从GHTorrent数据集中随机选择的，基于符合某些标准的项目的分层抽样。我们根据不同贡献者的数量将GitHub项目分为四组。我们为14个数据集中的每个数据集开发了六个回归模型（即，10 Gerrit的基础上和4 GitHub的基础上），以确定代码接受，审查间隔，和代码审查参与不同的性别和性别中立的配置文件的developer.ResultsOur结果发现显着的性别偏见在代码接受的14个数据集中的13个，7个有利于男性，其余6个有利于女性。我们发现男性和女性在代码审查间隔方面存在显着差异，在三种情况下，女性遇到的延迟时间比男性长，而在七种情况下则相反。我们的研究结果表明，审稿人的选择是最具性别偏见的方面之一，14个数据集中有12个表现出偏见。在14个案例中，共有11个案例显示，女性的代码审查参与率明显低于男性同事。由于大多数评审任务都是基于邀请，因此这一结果表明开发人员之间可能存在亲和力偏见。我们还注意到，女性使用性别中立档案的可能性明显更高。支持特雷尔等人“在Gerrit的三个项目中，性别中立的女性比男性接受代码的几率更高。然而，与他们的结果相矛盾的是，我们发现在所有四个GitHub项目组中，性别中立的女性接受代码的几率显着降低。类似的偏见缓解策略可能并不适用于所有社区，因为偏见的特点及其根本原因各不相同。由于邀请妇女参加审查的可能性较小，自由和开放源码软件项目应采取主动行动，确保公平地挑选妇女担任审查员。
ContextContemporary software development organizations lack diversity, and the ratios of women in Free and Open-Source Software (FOSS) communities are even lower than the industry average. Although the results of recent studies hint at the existence of biases against women, it is unclear to what extent such biases influence the outcomes of various software development tasks.ObjectiveThis study conceptually replicates two recent studies by Terrell et al. and Bosu and Sultana that investigated gender biases in FOSS communities. We aimto identify whether the outcomes of or participation in code reviews (or pull requests) are influenced by the gender of a developer.In particular, we focus on two outcome aspects (i.e., code acceptance, and review interval) and one participation aspect (i.e., code review participation) of code review processes.MethodWith this goal, this study includes a total of 1010 FOSS projects. Ten out of those projects use Gerrit-based code reviews. The remaining 1000 are randomly selected from the GHTorrent dataset based on a stratified sampling of projects fitting certain criteria. We divided GitHub projects into four groups based on the number of distinct contributors. We developed six regression models for each of the 14 datasets (i.e., 10 Gerrit based and 4 GitHub-based) to identify if code acceptance, review intervals, and code review participation differ based on the gender and gender-neutral profile of a developer.ResultsOur results find significant gender biases during code acceptance among 13 out of the 14 datasets, with seven favoring men and the remaining six favoring women. We found significant differences between men and women in terms of code review intervals, with women encountering longer delays than men in three cases and the opposite in seven. Our results indicate reviewer selection as one of the most gender-biased aspects, with 12 out of 14 datasets exhibiting bias. A total of 11 out of the 14 cases show women having significantly lower code review participation than their men colleagues. Since most of the review assignments are based on invitations, this result suggests possible affinity biases among the developers. We also noticed a significantly higher likelihood of women using gender-neutral profiles. Supporting Terrell et al.’s claim, women with gender-neutral profiles had higher odds of code acceptance than men among three Gerrit-based projects. However, contradicting their results, we found significantly lower odds of code acceptance for women with gender-neutral profiles across all four GitHub project groups.ConclusionsThough gender bias exists among many projects, the direction and amplitude of that bias vary based on project size, community, and culture. Similar bias mitigation strategies may not work across all communities, as the characteristics of biases and their underlying causes differ. As women are less likely to be invited for reviews, FOSS projects should take initiatives to ensure the equitable selection of women as reviewers.